Suraj Bhardwaj

Research Project: YOLO-Based Real-Time Object Detection for Autonomous Driving in CARLA • Developed a fully automated CARLA data generation pipeline for autonomous driving research. • Created a custom labeled dataset using a semi-automated workdlow — GroundedSAM for initial annotations and Robodlow for manual redinement. • Trained and benchmarked 12 YOLO models (v8–v11), with best achieving mAP50-95 ≈ 0.59, and deployed a real-time inference pipeline in CARLA. • Tech Stack: Python, PyTorch, CARLA, Robodlow

KARLI & SALSA Project: • Trained VoxelNeXt on the PANDASET dataset for 3D object detection and tracking. • Trained multimodal models (CLIP-ViP, Omnivore) using RGB, IR, depth, EEG, and audio for sleep pattern recognition. • Designed & developed an internal Retrieval-Augmented Generation (RAG) application by integrating RAPTOR-based retrieval. • Set up and maintained a multi-GPU Ray cluster for scalable training and automated hyperparameter optimisation. • Tech Stack: PyTorch, Ray, LangChain, ChromaDB, Git

Title: Improved Driver Distraction Detection Using Self-Supervised Learning • Proposed Clustered Feature Weighting, a label-free dataloading algorithm using HDBSCAN to handle skewed real-world datasets. • Achieved +7.17% balanced accuracy in RGB to IR cross-modality generalisation using DINOv2 ViTB/ 14 vs supervised baselines. • Designed reproducible pipelines for cross-modality and cross-view evaluation on the Drive&Act dataset. • Tech Stack: Python, Ray, PyTorch, Vision Transformers, Scikit-learn, OpenCV

Focus: Deep Learning and Computer Vision, Programming(C++, OOP, Unit Testing), Software Engineering (SDLC, UML, Scrum, Python, SQL, Text Mining)

Focus: Engineering Mathematics, Fluid Mechanics, Data Structure, Finite Elements Methods

Subjects: Physics, Chemistry, Mathematics, Informatics Practices, English